Dual-band flame detection method and system based on visible light and near-infrared images

Abstract

The invention discloses a dual-band flame detection method and system based on visible light and near-infrared images. The dual-band flame detection method comprises the following steps: collecting a dual-band flame image, carrying out joint calibration on a visible light camera and infrared imaging equipment, and obtaining internal and external parameters of the visible light camera and the infrared imaging equipment; performing horizontal alignment on the collected visible light image and the infrared image by using distortion correction and epipolar correction; marking the visible light image and the near-infrared image after horizontal alignment, and constructing a dual-band flame data set; constructing a dual-band flame detection model based on visible light and near-infrared images; and training a dual-band flame detection model by using the dual-band flame data set, and monitoring flame information in the scene in real time by the trained dual-band flame detection model. Compared with an existing flame detection model, the detection precision is higher, and the anti-interference capability is higher.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent detection, and relates to intelligent flame detection, in particular to a dual-band flame detection method and system based on visible light and near-infrared images. BACKGROUND

[0002] Traditional flame image detection mainly relies on image processing techniques, which identify flames by analyzing color, shape, texture and other features. Common methods include color model method, edge detection method and background modeling method. The color model method distinguishes flames from other objects by setting specific color thresholds (such as red, yellow and other warm colors), but is easily affected by lighting and environmental changes; the edge detection method extracts unique edge features of flames for identification, which is suitable for dynamic flames, but has poor adaptability to complex backgrounds; the background modeling method detects flames by identifying dynamic regions different from the background, which is commonly used in video monitoring, but still faces the problem of insufficient detection accuracy in complex environments. In addition, these methods usually rely on manually set rules and thresholds, lack sufficient robustness, and are difficult to adapt to different types of flames or diverse scenes.

[0003] Target detection technology is widely used in various scene detection due to its high accuracy and speed. This method mainly detects and identifies target information in images in real time through deep neural networks. Currently, its application in flame and smoke detection can be divided into two categories: one-stage and two-stage. One-stage means one-step, which directly obtains the candidate region of the target through a convolutional neural network and performs classification and positioning; two-stage algorithm first generates candidate regions of the image, and then classifies and locates the predicted target through a convolutional neural network. Although flame detection target detection technology has made great progress, it has not been widely applied in reality. Limited by the information limitation caused by single-band imaging, in simple scenes, using only visible light images for detection may be sufficient, but in the presence of multiple light source interference, such as mobile phone light, sunlight on sunny days, car lights, etc., a single visible light sensor will be greatly affected, leading to inaccurate flame detection. These light sources are very similar to flames in the visible light wavelength range, making it difficult for the model to accurately distinguish between flames and background light sources.

[0004] Currently, most research on flame detection based on deep learning is limited to the visible light modality, and there are many unresolved problems in multi-modal fusion detection of flames. Many multi-modal fusion works for flame detection tend to fuse at the decision level, and a better fusion method has not been designed at the feature level. Only simple concat channel stacking or pixel-by-pixel addition is done, and the correlation of features in two modalities is not studied. SUMMARY

[0005] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a dual-band flame detection method and system based on visible light and near-infrared images. It adopts a dual-channel feature extraction paradigm and fuses the feature information of the two bands according to the correlation to construct a dual-band flame detection model based on visible light and near-infrared images, so as to improve the detection accuracy.

[0006] Therefore, the present invention adopts the following technical solution.

[0007] In a first aspect, the present invention provides a dual-band flame detection method based on visible light and near-infrared images, comprising:

[0008] Acquire dual-band flame images, which include visible light images provided by a visible light camera and near-infrared images provided by an infrared imaging device, with the visible light camera and the infrared imaging device capturing flame images at the same location;

[0009] By photographing the calibration board, the visible light camera and infrared imaging device are jointly calibrated to obtain the internal and external parameters of the visible light camera and infrared imaging device.

[0010] Based on the aforementioned internal and external parameters, distortion correction and epipolar correction are used to horizontally align the collected visible light images and infrared images.

[0011] A dual-band flame dataset is constructed by annotating horizontally aligned visible light and near-infrared images.

[0012] A dual-band flame detection model based on visible light and near-infrared images was constructed using the YOLOv5 target detection network.

[0013] A dual-band flame detection model is trained using a dual-band flame dataset. The trained dual-band flame detection model monitors flame information in the scene in real time.

[0014] Furthermore, the construction steps of the dual-band flame detection model are as follows:

[0015] Step 1: Use the YOLOv5 object detection model as the baseline model;

[0016] Step 2: Add a near-infrared flame feature extraction branch to the YOLOv5 target detection model so that it can extract visible light and near-infrared flame feature information simultaneously.

[0017] Step 3: Fusion is performed based on the correlation between visible light and near-infrared flame characteristic information.

[0018] The dual-band flame detection model of this invention mainly addresses two issues: First, it adds a near-infrared flame feature extraction branch, fusing visible light feature information for flame detection. Second, it solves the problem of low feature-level fusion granularity between the two modes in the dual-band flame detection network, fully utilizing information from both visible and near-infrared modes to improve the model's detection accuracy and reduce false alarms and other problems.

[0019] Furthermore, the Transformer module is used to fuse visible light and near-infrared flame data features at the feature level, enabling the dual-band flame detection model to distinguish spatial information between different markers during training. This allows the self-attention mechanism to learn the binary relationship between visible light and near-infrared modes, resulting in a more refined fusion granularity.

[0020] Furthermore, visible light cameras and infrared imaging devices simultaneously acquire images of the flames.

[0021] Furthermore, the Zhang Zhengyou calibration method was used to jointly calibrate the visible light camera and the infrared imaging equipment.

[0022] Furthermore, the calibration process is as follows: First, fix the camera and image the checkerboard calibration board with different rotation and pitch angles to capture calibration image pairs. Then, use the Stereo Camera Calibrator toolbox to calibrate the captured infrared images. Click Add Images to load the image pairs, select 3Coefficients, and select Skew and Tangential Distortion to calculate the skew coefficient and tangential distortion. Click Calibrate to calibrate the visible light camera and infrared imaging device, and obtain the intrinsic parameters M1 and distortion coefficient D1 of the visible light camera, the intrinsic parameters M2 and distortion coefficient D2 of the infrared imaging device, and the extrinsic parameters (R, T) of the visible light camera and infrared imaging device.

[0023] Furthermore, the specific steps for horizontal alignment of visible light and infrared images are as follows: Using the StereoRectify tool in the OpenCV toolbox, distortion correction and epipolar correction are performed on the intrinsic and extrinsic parameters of the visible light camera and the infrared imaging device to obtain unified intrinsic parameters. Then, initUndistortRectifyMap is used to generate a remapping map, and then remap is used to perform pixel shifting on the visible light image I1 and the near-infrared image I2 to obtain the corrected visible light image I1' and the corrected near-infrared image I2'.

[0024] Furthermore, the visible light camera is slidably mounted on the guide rail, and the infrared imaging device is also slidably mounted on the guide rail, adjacent to the visible light camera.

[0025] Secondly, the present invention provides a dual-band flame detection system based on visible light and near-infrared images, comprising:

[0026] Image acquisition unit: used to acquire dual-band flame images, the dual-band flame images including visible light images provided by a visible light camera and near-infrared images provided by an infrared imaging device, the visible light camera and the infrared imaging device capturing flame images at the same location;

[0027] Camera calibration unit: By photographing the calibration board, it performs joint calibration of the visible light camera and the infrared imaging device, and obtains the internal and external parameters of the visible light camera and the infrared imaging device.

[0028] Horizontal alignment unit: Based on the aforementioned intrinsic and extrinsic parameters, distortion correction and epipolar correction are used to horizontally align the collected visible light image and infrared image;

[0029] Dataset construction unit: Annotated horizontally aligned visible light and near-infrared images to construct a dual-band flame dataset;

[0030] Detection model construction unit: Construct a dual-band flame detection model based on visible light and near-infrared images using the YOLOv5 target detection network;

[0031] Real-time monitoring unit: A dual-band flame detection model is trained using a dual-band flame dataset. The trained dual-band flame detection model monitors flame information in the scene in real time.

[0032] The present invention has the following beneficial effects:

[0033] 1) This invention differs from previous fire detection models such as SSD, Faster-RCNN, YOLOv3, and YOLOX by adopting a dual-channel feature extraction paradigm. It adds a near-infrared fire source feature extraction branch to the YOLOv5 model, enabling it to extract visible light and near-infrared flame feature information simultaneously. This effectively enhances the model's anti-interference ability and improves its recognition accuracy.

[0034] 2) Compared with the detection networks in the background technology that only use visible light or infrared light, or networks that train two networks separately and fuse them at the decision layer for fire detection, the fusion granularity of the present invention is more refined. In response to the misalignment problem between near-infrared and visible light, the fusion is performed based on the correlation of the feature information of the two bands. Transformer position embedding enables dual-band flame detection to distinguish the spatial information between different markers during training, and then a self-attention mechanism can be used to learn the binary relationship between visible light and near-infrared modes. Attached Figure Description

[0035] To more clearly illustrate the technical solution of the present invention, the drawings used in the prior art description will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention.

[0036] Figure 1 This is a flowchart of the dual-band flame detection method of the present invention;

[0037] Figure 2 This is a schematic diagram of the data acquisition device of the present invention;

[0038] Figure 3 This is a schematic diagram of distortion correction and polarity correction in this invention;

[0039] Figure 4 This is a schematic diagram of the dataset annotation for this invention;

[0040] Figure 5 This is a schematic diagram of the YOLOv5 model of the present invention;

[0041] Figure 6 This is a schematic diagram of the dual-channel feature extraction network of the present invention;

[0042] Figure 7 This is a schematic diagram of the dual-channel feature extraction and feature fusion network of the present invention;

[0043] Figure 8 The image shows the detection results of the dual-band fire source detection method based on visible light and near-infrared images of this invention.

[0044] Figure 9 This is a diagram showing the composition of the dual-band flame detection system of the present invention;

[0045] Figure 2 In the diagram, 1-visible light camera, 2-infrared imaging equipment, 3-rack, 4-server, B-horizontal distance between the two devices. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1

[0048] This embodiment provides a dual-band flame detection method based on visible light and near-infrared images, such as... Figure 1 As shown, the steps are as follows:

[0049] 1) Acquire dual-band flame images, including visible light images provided by a visible light camera and near-infrared images provided by an infrared imaging device. The infrared imaging device is placed next to the visible light camera, and both cameras capture flame images from the same location. The infrared imaging device 2 and the visible light camera 1 are fixed on a square frame 3 to form a dual-channel camera for acquiring dual-band flame images, and connected to a server 4. Figure 2 As shown.

[0050] 2) By photographing the calibration board, the visible light camera and infrared imaging equipment are jointly calibrated using the Zhang Zhengyou calibration method to obtain the internal and external parameters of the visible light camera and infrared imaging equipment.

[0051] The calibration process is as follows: First, fix the camera and image the checkerboard calibration board with different rotation and pitch angles to capture calibration image pairs. Then, use the Stereo Camera Calibrator toolbox to calibrate the captured infrared images. Click AddImages to load the image pairs, select 3Coefficients, and select Skew and Tangential Distortion to calculate the skew coefficient and tangential distortion. Click Calibrate to calibrate the visible light camera and infrared imaging device, and obtain the intrinsic parameters M1 and distortion coefficient D1 of the visible light camera, the intrinsic parameters M2 and distortion coefficient D2 of the infrared imaging device, and the extrinsic parameters (R, T) of the visible light camera and infrared imaging device.

[0052] The purpose of this step is to calibrate the visible light camera and infrared imaging equipment by photographing the calibration board, and to obtain the internal and external parameters of the visible light camera and infrared imaging equipment, so as to provide corresponding parameters for subsequent distortion correction and epipolar correction.

[0053] 3) Based on the aforementioned intrinsic and extrinsic parameters, distortion correction and epipolar correction are used to horizontally align the collected visible light and infrared images, such as... Figure 3 As shown.

[0054] Using the calibrated camera intrinsic and extrinsic parameters, the StereoRectify tool in the OpenCV toolbox is used to perform distortion correction and epipolar correction on the intrinsic and extrinsic parameters of the visible light camera and infrared imaging device to obtain unified intrinsic parameters. Then, initUndistortRectifyMap is used to generate a remapping map, and then remap is used to move pixels of the visible light image I1 and the infrared image I2 to obtain the corrected images I1' and I2'. This completes the horizontal alignment of the visible light image and the infrared image.

[0055] The purpose of this step is to use the intrinsic and extrinsic parameters and distortion coefficients of the visible light camera and the infrared imaging device to perform distortion correction and epipolar correction on the images captured by the two devices. This eliminates distortion caused by the camera lens and ensures that the images captured by the two devices are horizontally aligned, thereby reducing the error in the matching calculation in the dual-band flame detection model and improving the accuracy of the prediction results of the dual-band flame detection model.

[0056] 4) Label the horizontally aligned visible light and near-infrared images to construct a dual-band flame dataset.

[0057] like Figure 4 As shown, the visible light and infrared images, after manual annotation and horizontal alignment, follow standard procedures: Generally, the labeling software labelimg is used to annotate the images. The annotation dataset format is PascalVOC, with a single labeling category of "fire." A total of 3000 images are labeled as the dual-band flame dataset in this invention. The purpose of this step is to improve the quality of the dual-band flame dataset by annotating the images and selecting high-quality images, thereby providing high-quality training samples for the dual-band flame detection model and improving the model's flame detection accuracy.

[0058] 5) Construct a dual-band flame detection model based on visible light and near-infrared images using the YOLOv5 target detection network.

[0059] Specifically, the construction steps of the dual-band flame detection model are as follows:

[0060] Step 1: Use the YOLOv5 object detection model as the baseline model, such as... Figure 5 As shown, YOLOv5 is an end-to-end object detection algorithm based on deep learning. While maintaining high accuracy, this algorithm further improves detection speed and model lightweighting, and has broad prospects in practical applications.

[0061] Step 2: Add a near-infrared flame feature extraction branch to the YOLOv5 target detection model so that it can extract visible light and near-infrared flame feature information simultaneously.

[0062] Step 3: Fusion is performed based on the correlation between visible light and near-infrared flame characteristic information.

[0063] The Transformer module is used to fuse visible light and near-infrared flame data features at the feature level, enabling the dual-band flame detection model to distinguish spatial information between different markers during training. Furthermore, a self-attention mechanism can be used to learn the binary relationship between visible light and near-infrared modes, resulting in a more refined fusion granularity.

[0064] like Figure 6As shown, the main modification to the YOLOv5 model's Backbone section is the addition of a near-infrared flame feature extraction branch, whose structure is consistent with the visible light flame feature extraction branch. After capturing richer flame semantic information in the second, third, and fourth C3 modules of the feature extraction section, the visible light and near-infrared channel information are fused using Concat feature fusion. The fused feature information is then fed into the Neck section for feature processing, and finally, the prediction output is performed in the Head section.

[0065] like Figure 7 As shown, modifications were made to the backbone part based on the YOLOv5 model. First, a near-infrared flame feature extraction branch was added, with a structure consistent with the visible light flame feature extraction branch. Second, since the C3 module can capture richer semantic information of visible and near-infrared flames, a Transformer module was used after each C3 module in the backbone to deeply fuse visible and near-infrared feature information, and the fused feature information was added to the original mode branch, providing richer information for feature extraction in the subsequent backbone part and feature fusion in the neck part. Finally, to reduce computation, the feature map was sampled to a low resolution of H=W=8 before being passed to the Transformer module, and the output was upsampled to the original resolution through bilinear interpolation.

[0066]

[0067] In the formula a i,j This represents the correlation between the i-th and j-th positions on the feature map. When calculating the correlation matrix 'a', four matrix blocks can be naturally inferred. Two of these are intra-modal correlation matrix blocks (RGB and Infra), and the other two are inter-modal correlation matrix blocks. Therefore, with the help of the Transformer, it is only necessary to simply concatenate the multimodal features into a sequence, and the Transformer can automatically perform intra-modal and inter-modal information fusion simultaneously, robustly capturing the potential interactions between visible and near-infrared light.

[0068] 6) Train a dual-band flame detection model using a dual-band flame dataset. The trained dual-band flame detection model monitors flame information in the scene in real time.

[0069] like Figure 8As shown, the dual-band flame detection method based on visible light and near-infrared images is applied to practical detection. In nighttime environments, the YOLOv5 algorithm is easily interfered with by complex light sources such as vehicle headlights and streetlights, leading to an increased false alarm rate and misidentifying non-flame light sources as flames. In daytime strong light environments, the YOLOv5 algorithm faces several significant problems: firstly, it often fails to effectively detect weak and pale flames; secondly, it may misidentify similar-colored elements in the environment, such as sunlight shadows, yellow road dashed lines, and yellow leaves, as flames. These problems severely affect the accuracy and reliability of detection. In contrast, the method of this invention, by innovatively fusing image information from the visible light and near-infrared bands, significantly improves the accuracy and stability of flame detection. At night, the method of this invention, thanks to the advantage of dual-band fusion, effectively eliminates the influence of interfering light sources such as vehicle headlights and streetlights, avoiding false alarms; during the day, the method of this invention can accurately distinguish flames from other objects of similar color, successfully detecting weak, pale flames while reducing the false alarm rate.

[0070] In summary, the method proposed in this invention significantly improves the accuracy and reliability of flame detection under complex lighting conditions by optimizing the image information fusion strategy, and effectively reduces false alarms about flames.

[0071] Example 2

[0072] This embodiment provides a dual-band flame detection system based on visible light and near-infrared images to implement the dual-band flame detection method described in Embodiment 1, such as... Figure 9 As shown, it consists of an image acquisition unit, a camera calibration unit, a horizontal alignment unit, a dataset construction unit, a detection model construction unit, and a real-time monitoring unit.

[0073] Image acquisition unit: used to acquire dual-band flame images, the dual-band flame images including visible light images provided by a visible light camera and near-infrared images provided by an infrared imaging device, the visible light camera and the infrared imaging device capturing flame images at the same location;

[0074] Camera calibration unit: By photographing the calibration board, it performs joint calibration of the visible light camera and the infrared imaging device, and obtains the internal and external parameters of the visible light camera and the infrared imaging device.

[0075] Horizontal alignment unit: Based on the aforementioned intrinsic and extrinsic parameters, distortion correction and epipolar correction are used to horizontally align the collected visible light image and infrared image;

[0076] Dataset construction unit: Annotated horizontally aligned visible light and near-infrared images to construct a dual-band flame dataset;

[0077] Detection model construction unit: Construct a dual-band flame detection model based on visible light and near-infrared images using the YOLOv5 target detection network;

[0078] Real-time monitoring unit: A dual-band flame detection model is trained using a dual-band flame dataset. The trained dual-band flame detection model monitors flame information in the scene in real time.

[0079] It should be noted that each unit in the aforementioned dual-band flame detection system based on visible light and near-infrared images can be implemented entirely or partially through software, hardware, or a combination thereof. These units can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each unit. For specific limitations regarding the dual-band flame detection system based on visible light and near-infrared images, please refer to the limitations of the dual-band flame detection method based on visible light and near-infrared images (i.e., Embodiment 1) above; both have the same function and effect, and will not be repeated here.

[0080] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A dual-band flame detection method based on visible light and near-infrared images, characterized in that, include: Acquire dual-band flame images, which include visible light images provided by a visible light camera and near-infrared images provided by an infrared imaging device, with the visible light camera and the infrared imaging device capturing flame images at the same location; By photographing the calibration board, the visible light camera and infrared imaging device are jointly calibrated to obtain the internal and external parameters of the visible light camera and infrared imaging device. Based on the aforementioned internal and external parameters, distortion correction and epipolar correction are used to horizontally align the collected visible light images and infrared images. A dual-band flame dataset is constructed by annotating horizontally aligned visible light and near-infrared images. A dual-band flame detection model based on visible light and near-infrared images was constructed using the YOLOv5 target detection network. A dual-band flame detection model is trained using a dual-band flame dataset. The trained dual-band flame detection model monitors flame information in the scene in real time.

2. The dual-band flame detection method according to claim 1, characterized in that, The steps for constructing the dual-band flame detection model are as follows: Step 1: Use the YOLOv5 object detection model as the baseline model; Step 2: Add a near-infrared flame feature extraction branch to the YOLOv5 target detection model so that it can extract visible light and near-infrared flame feature information simultaneously. Step 3: Fusion is performed based on the correlation between visible light and near-infrared flame characteristic information.

3. The dual-band flame detection method according to claim 2, characterized in that, The Transformer module is used to fuse visible light and near-infrared flame data features at the feature level, and a self-attention mechanism is used to learn the binary relationship between visible light and near-infrared modes.

4. The dual-band flame detection method according to claim 1, characterized in that, Visible light cameras and infrared imaging devices simultaneously acquire images of the flames.

5. The dual-band flame detection method according to claim 1, characterized in that, The Zhang Zhengyou calibration method was used to jointly calibrate the visible light camera and the infrared imaging equipment.

6. The dual-band flame detection method according to claim 5, characterized in that, The calibration process is as follows: First, fix the camera and image the checkerboard calibration board with different rotation and pitch angles to capture calibration image pairs. Then, use the Stereo Camera Calibrator toolbox to capture the infrared images. Click AddImages to load the image pairs, select 3Coefficients, and select Skew and Tangential Distortion to calculate the skew coefficient and tangential distortion. Click Calibrate to calibrate the visible light camera and infrared imaging device, and obtain the intrinsic parameters M1 and distortion coefficient D1 of the visible light camera, the intrinsic parameters M2 and distortion coefficient D2 of the infrared imaging device, and the extrinsic parameters (R, T) of the visible light camera and infrared imaging device.

7. The dual-band flame detection method according to claim 6, characterized in that, The specific steps for horizontal alignment of visible light and infrared images are as follows: Using the StereoRectify tool in the OpenCV toolbox, distortion correction and epipolar correction are performed on the intrinsic and extrinsic parameters of the visible light camera and the infrared imaging device to obtain unified intrinsic parameters. Then, initUndistortRectifyMap is used to generate a remapping map, and then remap is used to move pixels of the visible light image I1 and the near-infrared image I2 to obtain the corrected visible light image I1' and the corrected near-infrared image I2'.

8. The dual-band flame detection method according to claim 1, characterized in that, The visible light camera is slidably mounted on the guide rail, and the infrared imaging device is also slidably mounted on the guide rail, next to the visible light camera.

9. A dual-band flame detection system based on visible light and near-infrared images, characterized in that, include: Image acquisition unit: used to acquire dual-band flame images, the dual-band flame images including visible light images provided by a visible light camera and near-infrared images provided by an infrared imaging device, the visible light camera and the infrared imaging device capturing flame images at the same location; Camera calibration unit: By photographing the calibration board, it performs joint calibration of the visible light camera and the infrared imaging device, and obtains the internal and external parameters of the visible light camera and the infrared imaging device. Horizontal alignment unit: Based on the aforementioned intrinsic and extrinsic parameters, distortion correction and epipolar correction are used to horizontally align the collected visible light image and infrared image; Dataset construction unit: Annotated horizontally aligned visible light and near-infrared images to construct a dual-band flame dataset; Detection model construction unit: Construct a dual-band flame detection model based on visible light and near-infrared images using the YOLOv5 target detection network; Real-time monitoring unit: A dual-band flame detection model is trained using a dual-band flame dataset. The trained dual-band flame detection model monitors flame information in the scene in real time.

10. The dual-band flame detection system according to claim 9, characterized in that, The steps for constructing the dual-band flame detection model are as follows: Step 1: Use the YOLOv5 object detection model as the baseline model; Step 2: Add a near-infrared flame feature extraction branch to the YOLOv5 target detection model so that it can extract visible light and near-infrared flame feature information simultaneously. Step 3: Fusion is performed based on the correlation between visible light and near-infrared flame characteristic information.

Images